EP1642087B1

EP1642087B1 - Method and apparatus for automatically detecting and mapping, particularly for burnt areas without vegetation

Info

Publication number: EP1642087B1
Application number: EP04745201A
Authority: EP
Inventors: Fabrizio Ferrucci; Barbara Rosalie Hirn
Original assignee: Individual
Current assignee: Individual
Priority date: 2003-07-09
Filing date: 2004-07-07
Publication date: 2006-12-27
Anticipated expiration: 2024-07-07
Also published as: WO2005005926A1; EP1642087A1; ITRM20030336A1; ITRM20030336A0; DE602004003944D1; ATE349677T1

Abstract

The invention concerns an automatic method of detecting and mapping, particularly for burnt areas without vegetation, which is based on the multi-temporal analysis of multi spectral data, or on the changes in the spectral response of the Earth features in combined near infrared, or NIR (Near Infra-Red), and short wave infrared, or SWIR (Short Wave Infra-Red), or medium infrared, or MIR (Medium Infra-Red), windows of the electromagnetic spectrum. The invention also concerns an electronic apparatus executing such method.

Description

The present invention refers to an automatic method of detecting and mapping, particularly for burnt areas without vegetation, enabling, in a precise, simple, versatile, reliable, and efficient way, the discrimination of areas, in images acquired from satellite, where fires occurred on the basis of a comparison of images overcoming possible problems of dishomogeneity arising in the case when the conditions of acquisition of the images to be compared are different. More in particular, the automatic mapping method is based on the multi-temporal analysis of multi spectral data, or on the changes in the spectral response of the Earth features in combined near infrared, or NIR (Near Infra-Red), and short wave infrared, or SWIR (Short Wave Infra-Red), or medium infrared, or MIR (Medium Infra-Red), windows of the electromagnetic spectrum.
The present invention further refers to an apparatus performing such a method, which is simple, fast, and inexpensive.
A spaceborne hot temperature event detection arrangement is described in document EP-A-1 048 928.
Multi spectral radiometers are generally constituted by one or more passive optical sensors that register the radiance emitted by the Earth in various windows of the electromagnetic spectrum.
The principal windows of the electromagnetic spectrum commonly used for land and vegetation analysis are the Visible (with wavelength from 0.4 µm to 0.7 µm, hereinafter referred to as VIS), the Near Infra-Red (with wavelength from 0.7 µm to 1.3 µm, hereinafter referred to as NIR), the Short Wavelength Infra-Red (with wavelength from 1.3 µm to 2.5 µm, hereinafter referred to as SWIR), the Medium Infra-Red (with wavelength from 2.5 µm to 4.5 µm, hereinafter referred to as MIR), and the Thermal Infra-Red (with wavelength up to 14.5 µm, hereinafter referred to as TIR).
Radiometers operating at these wavelengths are installed on airborne platforms - such as microlights, airplanes, balloons and dirigibles, operating between the ground free surface and the stratosphere - and spaceborne platforms, including all type of orbiting and sub-orbital spacecrafts at altitudes up to about 36.000 Km above the Earth's ground free surface.
All the afore-said systems look at the Earth orthogonally to the direction of the platform's motion with an angle ranging from 0 to 35 degrees with respect to Nadir - that is, perpendicularly to the Earth. The aforementioned angle is called "Off Nadir".
With such system, resolution cells or grids of the Earth surface can be generated with a spatial resolution (that is, the spacing of the elementary area, or pixel, within which it is possible to distinguish two objects) ranging from few centimetres to several kilometres.
The main characteristic of multi spectral radiometers is that they are coherent imaging systems. This means that they allow measuring radiated energy of each Earth surface cell at different wavelengths, as well as measuring the difference in radiated energy between two independent cells in each wavelength.
In theory, converting radiance in reflectance allows measuring the time variation in energy at each wavelength and in each Earth surface cell, and measuring the time variation of energy difference between two independent cells at each wavelength.
Considering an uniform Lambertian surface, and a cloudless atmosphere, and neglecting the fraction of the upward radiation backscattered downwards again by the atmosphere, the surface reflectance value p is related to radiance L_sen measured at the sensor by the following equation: $ρ = π (L_{sen} - L_{p}) / [d^{2} T_{v} (T_{z} E_{0} \cos (θ_{z}) + E_{d})]$
In equation [1], Lp is the path radiance (that is the additional radiance due to the path in the atmosphere and hence to the diffusion or scattering phenomenon), E_d is the downward irradiance, E₀ is exo-atmospheric solar irradiance, and θ_z is the solar zenith angle. The atmospheric transmittance T_v from the target toward the sensor depends on the off-nadir angle. The atmospheric transmittance T_z in the illumination direction is a function of θ_z, and of the optical thickness to be accounted for in Rayleigh scattering and atmospheric aerosol effects. Finally, d is the radius vector in astronomic units, given by $d = 1 / \{1 - 0.016729 * \cos [0.9856 (DOY - 4)]\},$
where DOY is the Day Of the Year.
Radiance Lp can be evaluated by using the so-called Dark Object Subtraction or DOS method, which, as shown by P.S. Chavez in "An improved Dark-Object Subtraction Technique for Atmospheric Scattering Correction of Multispectral Data", Remote Sensing of the Environment 24, pp. 459-479, 1988, relies upon the hypothesis that some ground types - as some types of rocks and soils - behave as almost perfect "black bodies": in such cases, the associated reflectance measured at the satellite is almost entirely due to scattered path reflectance. Atmospheric radiative transfer codes (Lowtran, Modtran, Code-6S, etc.) make use of standard atmosphere models, or of available measures of vertical atmosphere profiles, for determining the downward irradiance E_d. Finally, the accuracy of atmospheric corrections for Rayleigh scattering and aerosol effects strictly depend upon the availability of data on atmosphere physical and chemical parameters at the precise time of the remote sensing image acquisition
In case of multi-temporal analysis, the fundamental key is to distinguish reflectance changes originated by illumination and atmospheric conditions, by sensor drift or by sensor differences, from reflectance changes due to decrease in biomass content, to chlorophyll absorption, to water content and to density in the vegetation cover. A method for separating changes due to intrinsic reflectance value of vegetation from changes due to other factors influencing the at-satellite reflectance, is that of identifying "pseudo-invariant" features to be used as "reference targets" in cross-rectification of reflectance in different images. Identification of "pseudo-invariant" features requires accurate analysis of the whole of remote sensing data involved in the multi temporal analysis, following a linear radiometric transformation and aimed to implement relative reflectance retrieval. Accuracy of radiometric rectification strongly depends on the selection of pseudo-invariant features or of reference targets.
A large number of algorithms and processings intended for mapping burnt areas starting from multi spectral remote sensing data were carried out on single date image analysis, on the basis of so-called Principal Components, band ratio or vegetation index, unsupervised or supervised classification clustering. However, use of a single remote sensing image for mapping burnt and de-vegetated areas does not allow these ones to be distinguished from urban areas, water surfaces, sparsely vegetated areas and rock outcrops.
Multi-temporal methods based on the evaluation of changes of the spectral response between NIR and SWIR channels seem advantageous, since fire is an agent of land cover alteration, and therefore change detection methods reduce the likelihood of confusion with non vegetated or low vegetated static land cover types, such as urban areas, and rock outcrops
Multi temporal analysis were performed through Principal Component analysis followed by maximum likelihood classification, band ratio or vegetation index as NDVI index (Normalized Difference Vegetation Index) variations. The reflectance variation in areas passing from pre-fire situation to post-fire situation is strictly related to the vegetation type and the originally present biomass content. If only changes in reflectance response were considered, confusion may be then made between variations due to fire damage and variations due to hydric stress or seasonal vegetation changes.
It is therefore an object of the present invention to provide an automatic method capable to significantly improve the unsteadiness, and hence the poor reliability, of present processing techniques, allowing detecting, measuring, and mapping burned and de-vegetated areas in a precise, simple, versatile, reliable, and efficient way.
It is specific subject matter of this invention an automatic method of detecting and mapping, particularly for burnt areas without vegetation, in N images A_i, with N ≥ 2, for i = 1, 2,..., N, obtained from data acquired along a certain time period on the same geographical area by means of one or more spectral passive radiometers on one or more detection frequency channels, each image A_i comprising a same plurality of resolution cells or pixels p_xy, the reflectance p of each one of said one or more channels being available for each image, characterised in that it comprises the following steps:

B. for each pair of successive images A_i and A _i+1, for i = 1, 2,..., N-1, assuming as Permanent Reflectors PRs the resolution cells p_xy which have optical and electromagnetic properties constant in both images;
C. for each pair of successive images A_i and A _i+1, for i = 1, 2,..., N-1, calculating the mean M_{PR_P_i} and the standard deviation σ_{PR_P_i} of one or more first quantities depending on the variation, from the i-th image A_i to the time successive (i+1)-th image A _i+1, of one or more second quantities depending on at least one reflectance ρ of the Permanent Reflectors PRs in said one or more channels;
D. calculating the difference (M_{PR_Pi} - σ_Pi ) between the mean M_{PR_P_i} and the standard deviation σ_{PR_P_i}, calculated in step C, for each one of said one or more first quantities;
E. calculating the mean values M_{PR_ρ} of said one or more second quantities, depending on at least one reflectance ρ of the Permanent Reflectors PRs in said one or more channels, in each one of the second images A _i+1 of the pairs of successive images (A_i and A _i+1), for i =1,2,..., N-1;
F. for each resolution cell p_xy, calculating said one or more first quantities, already considered in step C for the Permanent Reflectors PRs, depending on the variation, from the i-th image A_i to the time successive (i+1)-th image A _i+1, of said one or more second quantities depending on at least one reflectance ρ of the pixels p_xy in said one or more channels;
G. determining a first set S₁ of resolution cells p_xy of the image A _i+1, comprising the resolution cells p_xy of the image A _i+1 for which said one or more first quantities calculated in step F are lower, in each one of said one or more channels, than the difference (M_{PR_Pi} - σ_Pi ) calculated in step D for each corresponding quantity of said one or more first quantities related to the Permanent Reflectors PRs;
H. determining a second set S₂ of resolution cells p_xy of the image A _i+1 comprising the resolution cells p_xy of the image A _i+1 (for i = 1, 2,..., N-1) for which said one or more second quantities, depending on at least one reflectance ρ of the resolution cell p_xy in said one or more channels in each one of the second images A _i+1 of the pairs of successive images, for i = 1, 2,..., N-1, are lower than all the mean values M_{PR_ρ} of the corresponding one or more second quantities of the Permanent Reflectors PRs calculated in step E;
I. determining a third set S₃ of resolution cells p_xy of the image A _i+1 equal to the set resulting from the intersection of the first set S₁ with the second set S₂: $S_{3} = S_{1} \cap S_{2};$
J. determining a set DB of the resolution cells p_xy of the image A _i+1 classified as corresponding to zones which have been subject to a decrease in biomass content or to an increase in soil or mineral cover during the time interval between the acquisitions of the two successive images A_i and A _i+1, the set DB being equal to the set of the pixels p_xy belonging to the first set S₁ which do not belong to the third set S₃ : $D B = S_{1} - S_{3};$
L. determining, within the third set S₃, a fourth set S₄ ⊆ S₃ of resolution cells p_xy corresponding to areas provided with "burnable" vegetation in the first images A_i of the pairs of successive images, for i = 1, 2,..., N-1;
M. determining a set BA of the resolution cells p_xy of the image A _i+1 classified as corresponding to zones which have been subject to a fire during the time interval between the acquisition of the image A_i and the acquisition of the image A _i+1, the set BA being equal to the fourth set S₄: $B A = S_{4;}$
and
N. determining a set DV of the resolution cells p_xy of the image A _i+1 classified as corresponding to zones which have been subject to a loss of vegetation during the time interval between the acquisitions of the two successive images A_i and A _i+1, the set DV being equal to the set of pixels p_xy belonging to the second set S₂ which do not belong to the fourth set S₄ : $D V = S_{2} - S_{4} .$

Always according to the invention, said one or more detection frequency channels may be at least two infrared channels, preferably a near infrared or NIR (Near Infra Red) channel, and a short wave infrared or SWIR (Short Wave Infra Red) channel, and/or a medium infrared or MIR (Medium Infra Red) channel.
Still according to the invention, the method may preliminarily process the data acquired by said one or more spectral passive radiometers for obtaining said N images A_i, with N ≥ 2, for i = 1, 2,..., N, of the same geographical area having the same plurality of resolution cells or pixels p_xy and the same size thereof.
Furthermore according to the invention, in step B Permanent Reflectors PRs may be determined as those resolution cells p_xy which, in both images A_i and A _i+1 of the pair, have values of reflectance ρ in each one of said one or more channels which correspond to densely built urban areas, preferably presenting no or little vegetation with respect to the size of the resolution cells.
Always according to the invention, said one or more first quantities depending on a variation, considered in steps C, D, F, and G, may comprise at least one quantity identically equal to the corresponding variation.
Preferably according to the invention, said one or more first quantities depending on a variation, considered in steps C, D, F, and G, comprise at least one percentage of the corresponding variation.
Still according to the invention, said one or more second quantities depending on at least one reflectance ρ, considered in steps C, E, F, and H, may comprise at least one quantity identically equal to the corresponding reflectance ρ of a channel.
Furthermore according to the invention, said one or more second quantities depending on at least one reflectance ρ, considered in steps C, E, F, and H, may comprise at least one ratio between two reflectances ρ of two channels.
Always according to the invention, the acquired data may comprise the radiance L of each one of the N images A_i on each one of said one or more detection frequency channels, and the method, before executing step B, may transforms the radiance L into the reflectance ρ for each one of said one or more channels, preferably taking account of the exo-atmospheric solar irradiance E₀ and of the effects of the sun zenith angle θ_z, and comprising the estimation of the path radiance L_p, more preferably through the DOS method.
Preferably according to the invention, for determining the first set S₁ in step G and/or the second set S₂ in step H, the method takes vegetation seasonal differences into account.
Still according to the invention, the method may take vegetation seasonal differences into account through the use of one or more corrective factors, preferably depending on the solar elevation angles θ _{SEL_i} and θ _{SEL_(i+1)} and/or on their difference (θ _{SEL_i} - θ_{SEL_(i+1)}) at the acquisition instants of the two images A_i and A _i+1.
Preferably according to the invention, the method executes steps L, M, and N only if the following checking step gives a positive outcome:
K. checking that the two images A_i and A _i+1 of the pair have been acquired in periods of the year in which vegetation is in a comparable blooming condition,
while in the case step K gives a negative outcome the following step is executed:
P. determining a set DV of the resolution cells p_xy of the image A _i+1 classified as corresponding to zones which have been subject to a loss of vegetation during the time interval between the acquisitions of the two successive images A_i and A _i+1, the set DV being equal to the second set S₂: $D V = S_{2} .$
Furthermore according to the invention, step K may check whether the solar elevation angles θ_{SEL_i} and θ_{SEL_(i+1)} at the acquisition instants of the two images A_i and A _i+1 are both higher or both lower than a threshold value Θ, preferably equal to 35° (Θ= 35°).
It is further subject matter of this invention an Electronic apparatus for processing data acquired along a certain time period on the same geographical area by means of one or more spectral passive radiometers on one or more detection frequency channels, for supplying a mapping, in particular for burnt areas without vegetation, of N images A_i, with N ≥ 2, for i = 1, 2,..., N, each image A_i comprising a same plurality of resolution cells or pixels p_xy , the apparatus comprising:

a data processing unit;
at least one memory unit connected to said data processing unit; and
at least one input/output, or I/O, interface connected to said data processing unit,
the electronic apparatus being characterised in that it executes the previously illustrated automatic method of detecting and mapping, particularly for burnt areas without vegetation.

The present invention will be now described, by way of illustration and not by way of limitation, according to its preferred embodiments, by particularly referring to the Figures of the enclosed drawings, in which:

Figure 1 schematically shows a flow diagram of a preferred embodiment of the method according to the invention;
Figure 2 shows the spectral reflectance signature ρ_xy of 12 zones belonging to different land cover vegetation classes for two acquisition dates and the spectral response of the same zone after fire;
Figure 3 shows the spectral reflectance signature of some land cover classes of vegetation and the response of dense urban areas;
Figure 4 shows the spectral reflectance signatures of the areas corresponding to Figure 3 after having been damaged by fire and the spectral response of dense urban PR areas;
Figure 5 shows a first image of Mount Etna acquired by means of the multi spectral radiometer SPOT-4;
Figure 6 shows a second image of Mount Etna acquired by means of the multi spectral radiometer Landsat-7 ETM+;
Figure 7 shows a third image of Mount Etna obtained by means of the method of Figure 1 applied to the two images of Figures 5 and 6;
Figure 8 shows a fourth image of Mount Etna obtained by means of the method of Figure 1 applied to the two images of Figures 5 and 6;
Figure 9 shows a fifth image of Mount Etna obtained by means of the method of Figure 1 applied to the two images of Figures 5 and 6; and
Figure 10 shows a sixth image of Mount Etna obtained by means of the method of Figure 1 applied to the two images of Figures 5 and 6.

The method according to the invention performs a multi spectral measuring of burned and de-vegetated zones on the basis of N images A_i, with N ≥ 2, and for i = 1, 2,..., N, acquired over a certain time span by means of spectral passive radiometers. Such radiometers used I the method according to the invention have at least two Infra-red channels, among which a NIR channel and a SWIR or MIR channel.
process for multi spectral measuring of burned and de-vegetated zones which, with N ≥ 2 images acquired over an assigned time span, with multi spectral passive radiometers presenting at least two Infrared channels - one NIR and one SWIR or MIR - identifies for every resolution cell, the fire damaged ones in relationship to a pseudo invariant features or Permanent Reflectors PRs, characterized in that said fire damaged cells are identified through the following steps
In particular, for each pair of successive images A_i and A _i+1, for i = 1, 2,..., N-1, the method discriminates:
the set BA of resolution cells or pixels p_xy of the image A _i+1, identified by coordinates (xy), corresponding to zones which have been subject to fire during the time interval between the acquisition of image A_i and the acquisition of image A _i+1;
the set DB of resolution cells p_xy of the image A _i+1 corresponding to zones which have been subject to a decrease in biomass content or to an increase in soil or mineral cover during the time interval between the acquisitions of the two successive images A_i and A _i+1; and
the set DV of resolution cells p_xy of the image A _i+1 corresponding to zones which have been subject to a loss of vegetation during the time interval between the acquisitions of the two successive images A_i and A _i+1.
The method identifies the three sets BA, DB, and DV in relation to pseudo invariant features or Permanent Reflectors, also indicated hereinafter as "PRs", by performing the steps illustrated in the following with reference to Figure 1.
In the first step A, for each cell p_xy of each one of the N images A_i, radiance L_NIR of the NIR channel and radiance L_SWIR of the SWIR channel (and/or radiance L_MIR of the MIR channel) are transformed, respectively, into reflectance p_NIR and into reflectance ρSWIR (and/or in reflectance ρ_MIR ), respectively, preferably taking account of the exo-atmospheric solar irradiance E₀ and of the effects of the sun zenith angle θ_z, and estimating the path radiance L_p, more preferably through the DOS method, still more preferably by employing equation [1].
In the second step B, the resolution cells p_xy of the N images A_i, the values of reflectance ρ of which identify them as cells corresponding to densely built urban areas, presenting no or little vegetation, are assumed as PRs.
In the third step C, for each pair of successive images A_i and A _i+1, for i = 1, 2,..., N-1, some statistical parameters of percentage P of variation, from i-th image A_i to the time successive (i+1)-th image A _i+1, of reflectances ρ of the response of the Permanent Reflectors PRs are calculated. In particular, the following are calculated:
the mean M_{PR_P_NIR_i} of percentage P of variation of the reflectance ρ_NIR in NIR channel of the response of the Permanent Reflectors PRs, for all the pairs of successive images A_i and A _i+1, for i = 1,2,..., N-1 ;
the standard deviation σ_{PR_P_NIR_i} of percentage P of variation of the reflectance ρ_NIR in NIR channel of the response of the Permanent Reflectors PRs, for all the pairs of successive images A_i and A _i+1, for i = 1, 2,..., N-1 ;
the mean M _{PR_P_NIR/SWIR_i} (and/or the mean M _{PR_P_NIR/MIR_i}) of percentage P of variation of the ratio ρ_NIR / ρ_SWIR (and/or of the ratio ρ_NIR / ρ_MIR ) of the reflectances ρ_NIR and ρ_SWIR (and/or ρ_MIR ) in NIR/SWIR channel ratio (and/or in NIR/MIR channel ratio), for all the pairs of successive images A_i and A _i+1, for i = 1, 2,..., N-1 ; and
the standard deviation σ_{PR_P_NIR/SWIR_i} (and/or the standard deviation σ _{PR_P_NIR/SWIR_i}) of percentage P of variation of the ratio ρ_NIR / ρ_SWIR (and/or of the ratio ρ_NIR / ρ_MIR ) of the reflectances ρ_NIR and ρ_WEIR (and/or ρ _MIR ) in NIR/SWIR channel ratio (and/or in NIR/MIR channel ratio), for all the pairs of successive images A_i and A _i+1, for i = 1, 2,..., N-1.
In the fourth step D, it is calculated the difference $(M_{{PR}_{-} Pi} - σ_{Pi})$

between the values of the means M_{PR_Pi} and the corresponding standard deviation σ_Pi for the NIR channel and the NIR/SWIR channel ratio (and/or the NIR/MIR channel ratio), calculated in step C for each pair of successive images A_i and A _i+1, for i = 1, 2,...,N-1. Such difference gives an estimation of the maximum difference of the cumulative contribution in atmospheric transmittance T_v from the target towards the sensor, in atmospheric transmittance T_z in the illumination direction, in optical thickness t_R for Rayleigh scattering, and atmospheric aerosol effects E_A.
In the fifth step E, the mean values M_{PR_ρ} of reflectances ρ of the response of the Permanent Reflectors PRs in each one of the second images A _i+1 of the pairs of successive images (A_i and A _i+1), for i = 1, 2,..., N-1 are calculated. In particular, the following are calculated:
the mean M _{PR_ρNIR_(i+1)} of the reflectance ρ_NIR in NIR channel of the response of the Permanent Reflectors PRs in all the images A _i+1, for i = 1, 2,..., N-1; and
the mean M _{PR_ρNIR_ρSWIR_(i+1)} (and/or M _{PR_ρNIR_ρMIR_(i+1)} of the ratio ρ_NIR / ρ_SWIR (and/or of the ratio ρ_NIR / ρ _MIR) of the reflectances ρ_NIR and ρ_SWIR (and/or ρ_MIR ) in NIR/SWIR channel ratio (and/or in NIR/MIR channel ratio) of the response of the Permanent Reflectors PRs in all the images A _i+1, for i=1,2,...,N-1.
In the sixth step F, for each resolution cell p_xy, it is calculated the percentage P_xy of variation, from the i-th image A_i to the time successive (i+1)-th image A _i+1, of reflectances ρ, specifically: the percentage P _{xy_NIR_i} of variation of the reflectance ρ_NIR of pixel p_xy in NIR channel, and the percentage P _{xy_NIR/SWIR_i} (and/or P _{xy_NIR/MIR_i}) of variation of the ratio ρ_NIR / ρ_SWIR (and/or of the ratio ρ_NIR / ρ_MIR ) of the reflectances ρ_NIR and ρ_SWIR (and/or ρ_MIR ) of pixel p_xy in the NIR/SWIR channel ratio (and/or in the NIR/MIR channel ratio).
In the seventh step G, it is determined a first set S₁ of resolution cells p_xy of the image A _i+1 corresponding to zones which have been subject to a decrease in biomass content or an increase in soil or mineral cover during the time interval between the acquisitions of the two successive images A_i and A _i+1 .
In particular, the first set S₁ comprises the resolution cells p_xy of the image A _i+1 for which the percentages calculated in the sixth step F are lower than the difference (M_{PR_Pi}- σ _Pi ) calculated in the preceding step D in the corresponding channels of the response of the Permanent Reflectors PRs, calculated in the third step C. That is, the first set S₁ comprises the pixels p_xy for which the following inequalities are satisfied: $P_{x y_{-} N I R_{-} i} < (M_{P R_{-} P_{-} N I R_{-} i} - σ_{P R_{-} P_{-} N I R_{-} i}),$
and
${P_{{xy}_{-}}}_{NIR / {SWIR}_{-} i} < (M_{{PR}_{-} P_{-} NIR / {SWIR}_{-} i} - σ_{P R_{-} P_{-} NIR / {SWIR}_{-} i})$

$(and / {or \begin{matrix} \end{matrix} P}_{{xy}_{-} NIR / {MIR}_{-} i} < M_{{PR}_{-} P_{-} NIR / {MIR}_{-} i} - σ_{P R_{-} P_{-} NIR / {SWIR}_{-} i}) .$
In the eighth step H, it is determined a second set S₂ of resolution cells p_xy of the image A _i+1 corresponding to zones which have been subject to a loss of vegetation during the time interval between the acquisitions of the two successive images A_i and A _i+1.
In particular, the second set S₂ comprises the resolution cells p_xy of the image A _i+1 (for i = 1, 2,..., N-1) for which the reflectances ρ_xy in the various channels are lower than the mean value M_{PR_ρ} of reflectances p of the response of the Permanent Reflectors PRs in the corresponding channels of the corresponding images A _i+1, calculated in the fifth step E. That is, the second set S₂ comprises the pixels p_xy for which the following inequalities are satisfied: ${ρ_{{xy}_{-}}}_{{NIR}_{-} (i + 1)} < M_{{PR}_{-} {ρNIR}_{-} (i + 1)}$
${ρ_{{xy}_{-}}}_{{NIR}_{-} {SWIR}_{-} (i + 1)} < M_{{PR}_{-} {ρNIR}_{-} {ρSWIR}_{-} (i + 1)}$
$(and / {orρ}_{{xy}_{-} {NIR}_{-} {MIR}_{-} (i + 1)} < M_{{PR}_{-} {ρNIR}_{-} {ρMIR}_{-} (i + 1)}) .$
In the subsequent ninth step I, it is determined a third set S₃ of resolution cells p_xy of the image A _i+1 corresponding to zones which have been subject to a damage during the time interval between the acquisitions of the two successive images A_i and A _i+1.
In particular, the third set S₃ is equal to the set resulting from the intersection of the first set S₁ with the second set S₂: $S_{3} = S_{1} \cap S_{2}$
In the tenth step J, it is determined the set DB of the resolution cells p_xy of the image A _i+1 corresponding to zones which have been subject to a decrease in biomass content or to an increase in soil or mineral cover during the time interval between the acquisitions of the two successive images A_i and A _i+1. In particular, the set DB is equal to the set of the pixels p_xy belonging to the first set S₁ which do not belong to the third set S₃, (i.e. which do not belong to the second set S₂ ): $D B = S_{1} - S_{3}$
In the eleventh step K, a check is carried out for ascertaining whether the solar elevation angles θ_{SEL_i} and θ _{SEL_(i+1)} at the acquisition instants of the two images A_i and A _i+1 are both higher or both lower than a threshold value Θ.
Such check assures that the two images A_i and A _i+1 have been acquired in periods of the year in which vegetation is in a comparable blooming condition, in order to grant reliability of mapping. By way of example, in Italy it may be deemed that the two images A_i and A _i+1 have been both acquired in spring or summer if θ_{SEL_i} and θ_{SEL_(i+1)} are both higher than Θ= 35°, or they have been both acquired in autumn or winter if θ_{SEL_i} and θ _{SEL_(i+1)} are both lower than Θ = 35°. Such check may be easily carried out by checking the following inequality: $(θ_{{SEL}_{-} i} - Θ) * (θ_{{SEL}_{-} (i + 1)} - Θ) > 0$
Alternatively, in order to consider the vegetation seasonal differences, the method may comprise the use of corrective factors in calculations for determining the first set S₁ and/or the second set S₂, corrective factors preferably depending on the solar elevation angles θ_{SEL_i} and θ _{SEL_(i+1)}, and possibly on their difference (θ_{SEL_i} - θ _{SEL_(i+1)}), at the acquisition instants of the two images A_i and A _i+1.
In the case when the check of step K gives a positive outcome, the three steps L, M and N are performed; otherwise, if the check of step K gives a negative outcome, step P is performed.
In step L it is determined, within the third set S₃, a fourth set S₄ ⊆ S₃ of resolution cells p_xy corresponding to zones wherein values of reflectance ρ_xy of the preceding image A_i identify them as cells corresponding to areas provided with "burnable" vegetation, such as forestry, agro-forestry, and semi-natural areas, and zones of permanent crops (preferably no zone is subject to human changes).
In step M, it is determined the set BA of the resolution cells p_xy of the image A _i+1 corresponding to zones which have been subject to a fire during the time interval between the acquisition of the image A_i and the acquisition of the image A _i+1. In particular, the set BA is equal to the fourth set S₄ $BA = S_{4}$
In step N, it is determined the set DV of the resolution cells p_xy of the image A _i+1 corresponding to zones which have been subject to a loss of vegetation during the time interval between the acquisitions of the two successive images A_i and A _i+1. In particular, the set DV is equal to the set of pixels p_xy belonging to the second set S₂ which do not belong to the fourth set S₄ : $D V = S_{2} - S_{4}$
In the case when the check of step K gives a negative outcome, it is not possible to determine with sufficient reliability the set BA, and step P is performed, with which it is determined only the set DV of the resolution cells p_xy of the image A _i+1 corresponding to zones which have been subject to a loss of vegetation during the time interval between the acquisitions of the two successive images A_i and A _i+1. In particular, the set DV is equal to the second set S₂ : $D V = S_{2}$
Preferably, the method comprises a further final step wherein the pixels p_xy belonging to the set BA which are singly isolated from the other pixels of the set BA (or which belong to groups of pixels p_xy of the set BA creating a zone having a reduced surface extension) are re-assigned to the set to which the surrounding pixels belong. This is done in order to reduce wrong assignments due to occurrence of limit values of reflectance or of the related statistical parameters used in calculations performed by the method and which may introduce evaluation ambiguities. Similarly, in the case when pixels not assigned to the set BA are present within wide areas of pixels of the set BA, this final step of the method may re-assign these pixels to the set BA. Similar processings may be carried out for the sets DB and DV.
It is evident that the method according to the invention enables estimating biomass content and vegetation cover from one date to another and ascertaining presence of areas which have been subject to loss of vegetation or fires during the same period.
Characteristics and advantages of the method according to the invention are illustrated in the following with reference to Figures 2-10, which show some applications thereof.
Figure 2 shows the so-called spectral reflectance signature ρ_xy of 12 zones pertaining to different vegetation classes for two acquisition dates and the spectral response of the same zone after having burned. Data have been collected through satellite platform Landsat-5 TM. In Figure 2, Landsat 5 TM channels number 1-5, and 7 are indicated as 1-6 in X coordinate, while the corresponding reflectance percentage is reported in Y coordinate. The wavelength interval of channels 1-5, and 7 are respectively 0.45-0.52 (Visible, or VIS, Blue), 0.53-0.61 (VIS-Green), 0.63-0.69 (VIS-Red), 0.78-0.90 (NIR), 1.55-1.75 (SWIR) and 2.09-2.35 (SWIR)
The burned areas are obtained by a visual comparison of the September 12, 1997, and the August 30, 1998, Landsat 5 TM images, made by an operator, by using various RGB combination in order to visually enhance the vegetation alteration. Such areas, which are sites of fire events reported in National Forest Guards and/or National Fire Brigades databases, are selected as unequivocally identified shapes of fire-damaged surfaces. Vegetation classes are indicated as land use Corine classes (according to the nomenclature of the known CORINE European project of coordinated information on the land use in the European context: COoRdinated INformation on the european Environment). In particular, the vegetation classes presenting a minimum of three distinct burned areas with a surface greater than 3 hectares are considered as fire-damaged areas. Resolution cells p_xy situated in the inner central part of areas identified as fire-damaged have been selected and their associated reflectance percentages in channels 1-5, and 7 have been adopted as characteristic spectral signature of burnt area for the relevant vegetation class.
For each land cover Corine class, spectral signatures of vegetated resolution cells on the two Landsat 5 images acquired on two different dates are different. This means that a quantitative evaluation of the damages occurred from one date to another by using multi-temporal analysis of remote sensing data can be performed only if the vegetation variation is calculated in relationship to a reference. Intrinsic reflectance spectral signature of this reference target homogeneoc!sly distributed over the whole remote sensing image must be independent from the acquisition date of the remote sensing data. Such reference targets are just the Permanents Reflectors PRs. The analysis of the vegetation variation from one date to another must be calculated by considering the distance from the vegetation spectral variation to the PRs spectral variation. Thus, for each pixel the detection change from one date to another is decomposed in two components: one as representative of the variations due to difference of sensor sensitivity and radiometric range, to difference of wavelength centre value, to difference of up-welling and down-welling transmittance along the ground sensor path between the two image acquisition dates, and the second component as representative of the effective variation of the vegetation state between the two remote sensed image acquisition dates.
The reflectance variation from vegetation to burned area is strictly related to the vegetation type and biomass content originally present, whereas the spectral signature of fire-damaged areas, on the post-fire image, appears slightly depending on the initial vegetation cover.
In the NIR window, reflectance is directly proportional to the amount of biomass. Since green vegetation is very reflective in the NIR channel, the fire process typically gives rise to a reflectance decrease.
The NIR reflectance decrease associated to fire damage is particularly significant for vegetal land classes characterised by the presence of trees and/or shrubs, such as: vineyards, fruit trees and olive groves (respectively identified as classes 221, 222, and 223 in the CORINE classification) or agro-forestry areas, transitional woodland and shrubs (respectively identified as classes 244, 323 and 324). The poorly defined class 242, including all complex cultivation patterns, is also characterised by a notable NIR reflectance decrease.
For all vegetation classes that underwent this analysis, NIR reflectance sharply decreases with respect to SWIR reflectance after fire damage
Figure 3 shows the spectral reflectance signature in some Corine vegetation classes and the response from densely urbanised areas, always obtained in channels 1-5, and 7 of the satellite platform Landsat 5 TM radiometer. Said urban areas homogeneously distributed over the whole sensed image are considered as PRs. As in Figure 2, also in Figure 3 channels 1-5 and 7 of Landsat-5 TM are numbered as 1 to 6 in X coordinate.
It can be noted in Figure 3 that, except for the case of Corine class 333, corresponding to sparsely vegetated areas, and of Corine class 223, corresponding to olive groves, NIR reflectance percentage of PRs and reflectance percentage ratio in NIR/SWIR channels of PRs are lower than the corresponding values of the vegetation classes taken under consideration.
Figure 4. shows some spectral reflectance signatures of the fire-damaged areas corresponding to the preceding Corine classes, considered in Figure 3, and the spectral response of densely urbanised areas PR areas, still obtained in channels 1-5, 7 (indicated with numbers 1 to 6 in X coordinate) through the satellite platform Landsat 5 TM.
Figure 4 shows that fire-damaged areas have the same general form of spectra independently from the original vegetation cover class. Moreover, both NIR reflectance percentage of PRs and reflectance percentage ratio in NIR/SWIR channels of PRs are higher than the corresponding values of all fire-damaged areas, independently from the original vegetation cover class.
Figure 5 represents a false colour image of Mount Etna acquired by means of the satellite platform SPOT-4 in channels NIRSWIR-RED on 7th June 2001. SPOT-4 borne multi spectral radiometer is characterised by four channels in the following wavelength intervals: 0.50-0.59 (VIS-Green), 0.61-0.68 (VIS-Red), 0.78-0.89 (NIR) and 1.58-1.75 (SWIR). Pixels size of SPOT images is equal to 20 meters* 20 meters.
Figure 6 represents a false colour composite image of Mount Etna acquired by means of the satellite platform Landsat-7 in channels SWIR-NIR-RED on 29th July 2001. Landsat-7 borne radiometer pixels, originally having size equal to 30 meters*30 meters, have been resized to pixels of size equal to 20 meters*20 meters with re-sampling method that does not alter the original land radiative spectrum.
The subsequent Figures 7-10 show the results of an application of the method according to the invention in critical conditions with respect to better processing procedures for the method, according to the following reasons.
First of all, the two remote sensed images are acquired by two different multi-spectral radiometers, i.e. SPOT-4 (radiometer HRVIR-1) and Landsat-7 (radiometer ETM+), which are furthermore differen in radiometric response distribution, different in channel central wavelength, different in radiometric range, and different in pixel size.
Moreover, the two remote sensed images are acquired in different seasons since the pre-fire image has been acquired in late spring at the beginning of June, when the vegetation is characterised by a high level of biomass content and ground coverage, whereas the post-fire image has been acquired in summer at the end of July, when the vegetation is characterized by high hydric stress. Due to seasonal changes, all vegetation Corine classes are characterised by a substantial decrease in both the biomass content and the vegetation cover.
Finally, on the year 2001 Etna volcano was the site of eruptive activity during summer and particularly at the end of July. 2001 eruptive activity of Etna was characterised by high level of ash fall, that is a strong catalyst for vegetation cover alteration, since it contributes to the increase of mineral cover on the ground and to the local drop of chlorophyll transformation of biomass.
Figure 7 represents the image comprising resolution cells, of size equal to 20 meters*20 meters, classified through variation analysis as damaged vegetation cells, which are characterised by a decrease in percentage of biomass content (decrease in percentage of NIR reflectance value) and in vegetation cover (decrease in percentage of NIR/SWIR ratio) higher than the PRs spectral variation calculated between the two remote sensed images of 7th June 2001, acquired by SPOT-4, and 29th July 2001, acquired by Landsat 7 ETM+. In the case under consideration, PRs are dense urban areas, the spectral variation of which is assumed as representative of differences in sensor sensitivity and radiometric range, in wavelength centre value, or in response histogram in the wavelength interval of NIR and SWIR channels, in up-welling and down-welling transmittance along the ground sensor path between the two remote sensing images of 7th June 2001, acquired by SPOT-4, and 29th July 2001, acquired by Landsat 7 ETM+.
The North-West part of the image is mainly occupied by evergreen forests and it can be noted that for pixels classified as damaged resolution cells from 7th June to 29th July, the decrease in biomass content and vegetation cover is principally due to seasonal changes occurring between spring and summer vegetation states.
Differently, for pixels classified as damaged resolution cells from 7th June to 29th July in the Central East part of the image, the decrease in biomass content and vegetation cover is principally due to ash fall occurred during the eruption period of Etna in 2001.
Figure 8 shows all the de-vegetated resolution cells present in the Landsat 7 ETM+ image acquired on 29th July 2001. Said resolution cells are characterised by biomass content (given by the NIR reflectance value) and vegetation cover (given by NIR/SWIR ratio) lower than the one of PRs (dense urban areas) on final remote sensing data Landsat 7 ETM+ of 29th July. On this image, cold lava flows, eroded soils, rock areas and few urban sites are classified as de-vegetated resolution cells.
Figure 9 shows all the resolution cells wherein a decrease in biomass content and vegetation cover occurs from 7th June 2001 to 29th July 2001 and which are classified as de-vegetated resolution cells on 29th July 2001. In particular, in Figure 9 the pixel clusters constituted by more than 10 pixels selected by both multi-temporal variation (Figure 7) and post fire analysis (Figure 8) are classified as fire damaged areas.
Figure 10 represents the false colour composite image of 29th July 2001 acquired by Landsat-7 ETM + in channels SWIR-NIR-Blue overlaying the layer GIS of fire damaged surfaces larger than 1 hectare, defined in Figure 9.
The preferred embodiments have been above described and some modifications of this invention have been suggested, but it should be understood that those skilled in the art can make variations and changes, without so departing from the related scope of protection, as defined by the following claims.

Claims

Automatic method of detecting and mapping, particularly for burnt areas without vegetation, in N images A_i, with N ≥ 2, for i = 1, 2,..., N, obtained from data acquired along a certain time period on the same geographical area by means of one or more spectral passive radiometers on one or more detection frequency channels, each image A_i comprising a same plurality of resolution cells or pixels p_xy, the reflectance p of each one of said one or more channels being available for each image, characterised in that it comprises the following steps:
B. for each pair of successive images A_i and A _i+1, for i = 1, 2,..., N-1, assuming as Permanent Reflectors PRs the resolution cells p_xy which have optical and electromagnetic properties constant in both images;

C. for each pair of successive images A_i and A _i+1, for i = 1, 2,..., N-1, calculating the mean M_{PR_P_i} and the standard deviation σ_{PR_P_i} of one or more first quantities (P) depending on the variation, from the i-th image A_i to the time successive (i+1)th image A _i+1, of one or more second quantities (ρ _NIR , ρ_NIR / ρ_SWIR, ρ_NIR / ρ_MIR ) depending on at least one reflectance ρ of the Permanent Reflectors PRs in said one or more channels;

D. calculating the difference (M_{PR_Pi} - σ_Pi ) between the mean M_{PR_P_i} and the standard deviation σ_{PR_P_i}, calculated in step C, for each one of said one or more first quantities (P);

E. calculating the mean values M_{PR_ρ} of said one or more second quantities (ρ_NIR , ρ_NIR /ρ_SWIR, ρ_NIR / ρ_MIR ), depending on at least one reflectance ρ of the Permanent Reflectors PRs in said one or more channels, in each one of the second images A _i+1 of the pairs of successive images (A_i and A _i+1), for i = 1, 2,...,N-1 ;

F. for each resolution cell p_xy, calculating said one or more first quantities (P_xy ), already considered in step C for the Permanent Reflectors PRs, depending on the variation, from the i-th image A_i to the time successive (i+1)-th image A _i+1, of said one or more second quantities (ρNIR , ρNIR / ρSWIR, ρNIR / ρMIR) depending on at least one reflectance p of the pixels p_xy in said one or more channels;

G. determining a first set S₁ of resolution cells p_xy of the image A _i+1, comprising the resolution cells p_xy of the image A _i+1 for which said one or more first quantities (P_xy ) calculated in step F are lower, in each one of said one or more channels, than the difference (M_{PR_Pi} - σ_Pi ) calculated in step D for each corresponding quantity (P) of said one or more first quantities (P) related to the Permanent Reflectors PRs;

H. determining a second set S₂ of resolution cells p_xy of the image A _i+1 comprising the resolution cells p_xy of the image A _i+1 (for i = 1, 2,..., N-1) for which said one or more second quantities (ρNIR, ρNIR / ρSWIR, ρNIR /ρMIR), depending on at least one reflectance ρ of the resolution cell p_xy in said one or more channels in each one of the second images A _i+1 of the pairs of successive images, for i = 1, 2,..., N-1, are lower than all the mean values M_{PR_p} of the corresponding one or more second quantities (ρNIR, ρNIR / ρSWIR, ρNIR /ρMIR) of the Permanent Reflectors PRs calculated in step E;

I. determining a third set S₃ of resolution cells p_xy of the image A _i+1 equal to the set resulting from the intersection of the first set S₁ with the second set S ₂ : $S_{3} = S_{1} \cap S_{2};$

J. determining a set DB of the resolution cells p_xy of the image A _i+1 classified as corresponding to zones which have been subject to a decrease in biomass content or to an increase in soil or mineral cover during the time interval between the acquisitions of the two successive images A_i and A _i+1, the set DB being equal to the set of the pixels p_xy belonging to the first set S₁ which do not belong to the third set S₃ : $D B = S_{1} - S_{3};$

L. determining, within the third set S₃, a fourth set S₄ ⊆ S₃ of resolution cells p_xy corresponding to areas provided with "burnable" vegetation in the first images A_i of the pairs of successive images, for i = 1, 2,..., N-1;

M. determining a set BA of the resolution cells p_xy of the image A _i+1 classified as corresponding to zones which have been subject to a fire during the time interval between the acquisition of the image A_i and the acquisition of the image A _i+1, the set BA being equal to the fourth set S₄: $BA = S_{4};$
and

N. determining a set DV of the resolution cells p_xy of the image A _i+1 classified as corresponding to zones which have been subject to a loss of vegetation during the time interval between the acquisitions of the two successive images A_i and A _i+1, the set DV being equal to the set of pixels p_xy belonging to the second set S₂ which do not belong to the fourth set S₄ : $D V = S_{2} - S_{4} .$
Method according to claim 1, characterised in that said one or more detection frequency channels are at least two infrared channels.
Method according to claim 2, characterised in that said at least two infrared channels are a near infrared or NIR (Near Infra Red) channel, and a short wave infrared or SWIR (Short Wave Infra Red) channel, and/or a medium infrared or MIR (Medium Infra Red) channel.
Method according to any one of the preceding claims, characterised in that the method preliminarily processes the data acquired by said one or more spectral passive radiometers for obtaining said N images A_i, with N ≥ 2, for i = 1, 2,..., N , of the same geographical area having the same plurality of resolution cells or pixels p_xy and the same size thereof.
Method according to any one of the preceding claims, characterised in that in step B Permanent Reflectors PRs are determined as those resolution cells P_xy which, in both images A_i and A _i+1 of the pair, have values of reflectance ρ in each one of said one or more channels which correspond to densely built urban areas, preferably presenting no or little vegetation with respect to the size of the resolution cells.
Method according to any one of the preceding claims, characterised in that said one or more first quantities (P) depending on a variation, considered in steps C, D, F, and G, comprise at least one quantity identically equal to the corresponding variation.
Method according to any one of the preceding claims, characterised in that said one or more first quantities (P) depending on a variation, considered in steps C, D, F, and G, comprise at least one percentage (P) of the corresponding variation.
Method according to any one of the preceding claims, characterised in that said one or more second quantities (ρNIR, ρNIR / ρSWIR, ρNIR / ρMIR) depending on at least one reflectance ρ, considered in steps C, E, F, and H, comprise at least one quantity identically equal to the corresponding reflectance ρ (ρ_NIR ) of a channel.
Method according to any one of the preceding claims, characterised in that said one or more second quantities (ρNIR, ρNIR / ρSWIR, ρNIR /ρMIR) depending on at least one reflectance ρ, considered in steps C, E, F, and H, comprise at least one ratio between two reflectances p (ρNIR / ρSWIR, ρNIR / ρMIR) of two channels.
Method according to any one of the preceding claims, characterised in that the acquired data comprise the radiance L of each one of the N images A_i on each one of said one or more detection frequency channels, and in that the method, before executing step B, transforms ([1]) the radiance L into the reflectance ρ for each one of said one or more channels.
Method according to claim 10, characterised in that the transformation of the radiance L into the reflectance ρ takes account of the exo-atmospheric solar irradiance E₀ and of the effects of the sun zenith angle θ_z, and it comprises the estimation of the path radiance L_p, preferably through the DOS method.
Method according to any one of the preceding claims, characterised in that, for determining the first set S₁ in step G and/or the second set S₂ in step H, the method takes vegetation seasonal differences into account.
Method according to claim 12, characterised in that the method takes vegetation seasonal differences into account through the use of one or more corrective factors.
Method according to claim 13, characterised in that said one or more corrective factors depends on the solar elevation angles θ_{SEL_i} and θ_{SEL_(i+1)} and/or on their difference (θ_{SEL_i} - θ _{SEL_(i+1)}) at the acquisition instants of the two images A_i and A _i+1.
Method according to any one of claims 12 to 14, characterised in that it executes steps L, M, and N only if the following checking step gives a positive outcome:
K. checking that the two images A_i and A _i+1 of the pair have been acquired in periods of the year in which vegetation is in a comparable blooming condition,
while in the case step K gives a negative outcome the following step is executed:

P. determining a set DV of the resolution cells p_xy of the image A _i+1 classified as corresponding to zones which have been subject to a loss of vegetation during the time interval between the acquisitions of the two successive images A_i and A _i+1, the set DV being equal to the second set S₂ : $D V = S_{2} .$
Method according to claim 15, characterised in that step K checks whether the solar elevation angles θ_{SEL_i} and θ_{SEL_(i+1)} at the acquisition instants of the two images A_i and A _i+1 are both higher or both lower than a threshold value Θ.
Method according to claim 16, characterised in that said threshold value Θ is equal to 35° (Θ= 35°).
Electronic apparatus for processing data acquired along a certain time period on the same geographical area by means of one or more spectral passive radiometers on one or more detection frequency channels, for supplying a mapping, in particular for burnt areas without vegetation, of N images A_i, with N ≥ 2, for i = 1, 2,..., N , each image A_i comprising a same plurality of resolution cells or pixels p_xy, the apparatus comprising:
a data processing unit;

at least one memory unit connected to said data processing unit; and

at least one input/output, or I/O, interface connected to said data processing unit,

the electronic apparatus being characterised in that it executes the automatic method of detecting and mapping, particularly for burnt areas without vegetation, according to any one of claims 1 to 17.